[Ready To Merge][AQUA] Mutli-Model Deployment

[mrDzurb]

Description

Enhance AQUA SDK & CLI to Support Multi-Model Deployment

This PR introduces multi-model deployment support in the AQUA SDK and CLI, enabling users to deploy multiple models within a single model deployment. The enhancement includes updates to the SDK and CLI logic, allowing users to specify multiple models for deployment while ensuring proper validation and GPU resource allocation.

Key Changes

• Updated SDK AQUA Model Deployment API to support multiple models in a single deployment.
• Modified CLI commands to allow users to provide multiple model OCIDs when creating a deployment.
• Refactored model validation logic to ensure that:
  • Only LLM models compatible with the VLLM container can be deployed.
  • GPU allocation is properly handled across all selected models.
    • Implemented validation enhancements to check model compatibility with the selected shape and GPU allocation.
    • Updated SDK handlers to support retrieving and registering grouped models.
    • Improved metadata storage mechanism for tracking multiple models in a single deployment.
    • Unit tests added to validate multi-model functionality in the SDK and CLI.

Breaking Changes & Compatibility Considerations

• Existing single-model deployments remain unchanged, but the updated logic extends support for multi-model deployments.

Next Steps & Future Enhancements

• Enhance UI to reflect new SDK changes, allowing users to configure multi-model deployments via AQUA UI.
• Improve documentation to provide clear guidance on deploying multiple models in a single deployment.

Testing & Validation

• Unit tests updated to cover new multi-model deployment logic.
• Integration tests performed to validate CLI interactions with multi-model deployment APIs.
• Manual testing conducted to verify proper GPU allocation and deployment success.
  • The manual tests covered in the: [AQUA][MMD] E-2-E Tests confluence page.

CLI Test

List Available Shapes

ads aqua deployment list_shapes

Retrieve Deployment Configuration*

ads aqua deployment get_multimodel_deployment_config --model_ids '["ocid1.datasciencemodel.oc1.iad.<OCID>"]'

Create a Multi-Model Deployment

ads aqua deployment create --container_image_uri "dsmc://odsc-vllm-serving:0.6.4.post1.2" --display_name "multi-model-deployment-test-completeion" --instance_shape "BM.GPU.A10.4" --models '[{"model_id":"ocid1.datasciencemodel.oc1.iad.<OCID>", "gpu_count":2},{"model_id":"ocid1.datasciencemodel.oc1.iad.<OCID>", "gpu_count":2}]' --log_group_id "ocid1.loggroup.oc1.iad.<OCID>" --access_log_id  "ocid1.log.oc1.iad.<OCID>" --predict_log_id "ocid1.log.oc1.iad.<OCID>"

Get Details for Model Deployment

ads aqua deployment get --model_deployment_id "ocid1.datasciencemodeldeployment.oc1.iad.<OCID>"

[github-actions]

📌 Cov diff with main:

📌 Overall coverage:

[github-actions]

📌 Cov diff with main:

📌 Overall coverage:

[github-actions]

📌 Cov diff with main:

📌 Overall coverage:

[github-actions]

📌 Cov diff with main:

📌 Overall coverage:

[github-actions]

📌 Cov diff with main:

📌 Overall coverage:

[github-actions]

📌 Cov diff with main:

📌 Overall coverage:

[github-actions]

📌 Cov diff with main:

📌 Overall coverage:

[VipulMascarenhas]

added minor comments, changes look good.

[github-actions]

📌 Cov diff with main:

📌 Overall coverage:

[github-actions]

📌 Cov diff with main:

📌 Overall coverage:

[github-actions]

📌 Cov diff with main:

📌 Overall coverage: